def svrmodel(traininput, traintarget, testinput, testtarget):
# [points, label] = weeklydataset_shogun(filesource, [])
# clusteredpoints, cdata = create_clustered_samples(points, 5, 1)
# cluster0 = clusteredpoints[0]
# input, target = aggregatebymins_sg(cluster0[0])
## testinput, testtarget = aggregatebymins_sg(cluster0[0])
# print len(input)
# traininput, traintarget, testinput, testtarget = traintest(input, target, 20, 1)
svr = SVR(traininput, testinput, traintarget,2,256,0.1,0.5)
out = svr.svr_req(testinput)
sme = svr.calc_sme(testtarget, out)
x = array(testinput, dtype=int32)
y = array(testtarget, dtype=int32)
xp = array(testinput, dtype=int32)
yp = array(out, dtype=int32)
fig = figure()
ax1 = fig.add_subplot(1,1,1)
ax1.plot(x, y)
ax1.plot(xp,yp,"r")
ax1.axis([0,max(x)+10,0,max(y)+10])
ax1.set_xlabel('minutes of the week')
ax1.set_ylabel('number of requests')
fig.savefig("svr_model_%f" % (time.time()), format='png')
print "SME = %f" % sme
def svr(C, gamma, eps):
#initialization of data wmproxy
traininput, traintarget, testinput, testtarget = initialize_wmproxy()
#training of the SVR
#scaling values in training and test targets
for i in range(len(traintarget)):
if(traintarget[i] != 0):
traintarget[i] = log(traintarget[i])
if(traininput[i] != 0):
traininput[i] = log(traininput[i])
for i in range(len(testtarget)):
if(testtarget[i] != 0):
testtarget[i] = log(testtarget[i])
if(testinput[i] != 0):
testinput[i] = log(testinput[i])
avg = mean(traintarget)
sigma = std(traintarget)
maxtrain = len(traintarget)
C = max([abs(avg + sigma), abs(avg - sigma)])
print "C is equal to %f" % C
svr = SVR(traininput[maxtrain-1440:maxtrain], testinput, traintarget[maxtrain-1440:maxtrain],gamma,C,eps,eps)
out = svr.svr_req(testinput[0:30])
error = 0
for i in range(len(out)):
error += (out[i] - testtarget[i])
mean_error = error / len(out)
variance = 0
for i in range(len(out)):
variance = abs(out[i] - mean_error)
variance /= len(out)
print "Variance = %f" % variance
epsilon = 3*variance*sqrt(log(len(out))/len(out))
print "Epsilon = %f" % epsilon
#calculation of the metrics
sme = svr.calc_sme(testtarget[0:30], out)
mape = svr.calc_mape(out, testtarget[0:30])
predx = svr.calc_pred(out, testtarget[0:30], 25)
rsq = svr.calc_rsqr(out, testtarget[0:30])
print out
print testtarget[0:30]
# print model results!
x = array(testinput[0:30], dtype=int32)
y = array(testtarget[0:30], dtype=int32)
xp = array(testinput[0:30], dtype=int32)
yp = array(out, dtype=int32)
fig = figure()
ax1 = fig.add_subplot(1,1,1)
ax1.title.set_text("Predizioni modello SVR con C= %f, Gamma = %f, Eps = %f" % (C, gamma, eps))
realvalues = ax1.plot(x, y)
predictedvalues = ax1.plot(xp,yp,"r")
ax1.axis([8.9,max(xp)+0.5,0,max(y)+10])
ax1.set_xlabel('minutes of the week')
ax1.set_ylabel('number of requests')
legend([realvalues,predictedvalues], ["Real Values","Predicted Values"])
fig.savefig("svr_model_%f" % time(), format='png')
print "SME = %f" % sme
print "MAPE = %f" % mape
print "R^2 = %f" % rsq
print "PREDX = %f" % predx